AI

UCLA researchers’ new study says the GPT-3 language model can reason with analogy, but they can’t say the answers aren’t already in the training data. The latest in a slew of speculative AI research papers is making some pretty outlandish claims about how deep learning models have some subtle, unrealized cognitive abilities akin to, or even surpassing humans. Though researchers found a modern pre-training transformer model does well at multiple-choice tests that don’t necessarily require language, they still have no honest idea if the AI is just basing its answers off its opaque training data. University of California, Los Angeles researchers tested “analogical tasks” on the GPT-3 large language model and found it was at or above “human capabilities” for resolving complex reasoning problems. UCLA was fast to make rather outlandish claims about the research in its  press release  Monday, raising the question of whether the AI was “using a fundamentally new kind of cog...

 When Google had its unpleasant realization that it had been complacently spinning its wheels on a form of fake AI for a decade, chances are it started realigning itself that day. And it sounds like Assistant itself is now getting a generative face-lift, according to an internal email reported by Axios.   The email says that the Assistant team leads “see a huge opportunity to explore what a supercharged Assistant, powered by the latest LLM [large language model] technology, would look like,” and describe some organizational changes to achieve that.   Of course, you don’t make sweeping changes to a successful division just because you want to see what something looks like. It feels more like they have already seen what it looks like as other companies have demonstrated it publicly, and they are in a hurry to catch up. At any rate, the change in “vision” will unfold over the months to come.   Although there are numerous examples of LLMs powering chatbots and assistants...